[
https://issues.apache.org/jira/browse/FLINK-9565?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16524641#comment-16524641
]
yinhua.dai commented on FLINK-9565:
-----------------------------------
[~fhueske] [~twalthr]
So there are two options:
#1. using annotation like @Expensive
{code:java}
@Expensive
public class MyScalarFunc extends ScalarFunction {
public Double eval(Double d) {
// expensive logic
}
}{code}
#2. using additional method such as evalParallel, then sample code would be
{code:java}
public class MyScalarFunc extends ScalarFunction {
public Double evalParallel(Double d) {
// expensive logic
}
}
{code}
I personally prefer to option #1, I feel like it makes the code generation
easier so we always call "eval" method, and we simply check if the scalar
function is wrapped with @Expensive annotation, in which case we wrapped the
generated code within a future.
> Evaluating scalar UDFs in parallel
> ----------------------------------
>
> Key: FLINK-9565
> URL: https://issues.apache.org/jira/browse/FLINK-9565
> Project: Flink
> Issue Type: New Feature
> Components: Table API & SQL
> Affects Versions: 1.4.2
> Reporter: yinhua.dai
> Priority: Major
>
> As per
> [https://stackoverflow.com/questions/50790023/does-flink-sql-support-to-run-projections-in-parallel,]
> scalar UDF in the same SQL is always evaluated sequentially even when those
> UDF are irrelevant, it may increase latency when the UDF is time consuming
> function.
> It would be great if Flink SQL can support to run those UDF in parallel to
> reduce calculation latency.
>
> cc [~fhueske]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
